Damage and Loss Assessment of a Building Subjected to a Specific Ground Motion

[rezvan]

Hello Adam,

Thank you for the updated Version 3.b4 of Pelicun.

I have a question and I will be so grateful if you could help me.

Is it possible to get a more deterministic damage and loss assessment of a building when is subjected to a specific ground motion by setting the "log_std" of the demands equal to zero?

(I know the repair time and costs are also accompanied by uncertainty as well, so the final results will not be purely deterministic).

In the example you provided in the Jupiter notebook, I set the "log_std" of the demands equal to zero, but I observed meaningful differences in the results (compared to the non-zero "log_std" case).

I attached the summary tables of the consequences here and as you can see, the standard deviation of the consequences is even larger when the "log_std" is set to zero.

Thank you so much,

Pooya

[adamzs]

Hi Pooya,

I'm sorry for the delayed response.

If you want to have a deterministic demand, you can simply provide identical realizations in the demand_sample. There is no need to hack the system by specifying a lognormal distribution with almost zero log_std. In the first pelicun LET example, we first generate demands, then extend the sample and feed it back to pelicun. If you prepare your own sample, you can skip the first part, and start at the point where we feed the sample back. Let me know if you need further details on this.

As for the standard deviation in the results, I am not sure that is a meaningful statistic in this case. Based on what I see in the first table, you only have zero damage or collapse (i.e., total loss) there. The mean is around 1 million, while the total loss is 21 million. About 10% of the results at 21 million might lead to a large standard deviation on paper, but your data really is at zero and at that large value, so there is not much dispersion. I am not sure those results are valid, though. When you fed in zero log_std, that might have led to unexpected behavior. I recommend trying again with the demand sample as I mentioned above and checking how the results come out.

Finally, I wanted to mention that you can remove the uncertainty from the fragility and consequence functions as well if that is what you need to do for your work. Let me know if you are interested in that I am happy to tell you more details.

Adam

[rezvan]

Hi Adam,

I am thankful for your timely and detailed response.

Actually, after I set the "log_std" column to a very small value (e.g. 10^-6), I checked the generated realizations (attached), and there wasn't any problem in generating deterministic values of each EDP (attached photo) and fortunately, Pelicun was successful in generating the same EDP value in each realization.

However, I was not paying attention to the residual drift since in the example you provided in Jupyter Notebook, the empirical relationships were being used to predict the residual drift and therefore they were taking different values for each realization. I should use deterministic values for the residual drifts as well and see the difference.

Again, I appreciate your enlightening response.

Pooya,